Chapter 22
Practical Computational Methods for Regulatory Genomics: A cisGRN-Lexicon and cisGRN-Browser for Gene Regulatory Networks
Sorin Istrail, Ryan Tarpine, Kyle Schutter, and Derek Aguiar
Abstract
The CYRENE Project focuses on the study of cis-regulatory genomics and gene regulatory networks (GRN) and has three components: a cisGRN-Lexicon, a cisGRN-Browser, and the Virtual Sea Urchin software system. The project has been done in collaboration with Eric Davidson and is deeply inspired by his experimental work in genomic regulatory systems and gene regulatory networks. The current CYRENE cisGRN-Lexicon contains the regulatory architecture of 200 transcription factors encoding genes and 100 other regulatory genes in eight species: human, mouse, fruit fly, sea urchin, nematode, rat, chicken, and zebrafish, with higher priority on the first five species. The only regulatory genes included in the cisGRN-Lexicon (CYRENE genes) are those whose regulatory architecture is validated by what we call the Davidson Criterion: they contain functionally authenticated sites by site-specific muta- genesis, conducted in vivo, and followed by gene transfer and functional test. This is recognized as the most stringent experimental validation criterion to date for such a genomic regulatory architecture. The CYRENE cisGRN-Browser is a full genome browser tailored for cis-regulatory annotation and inves- tigation. It began as a branch of the Celera Genome Browser (available as open source at http:// sourceforge.net/projects/celeragb/) and has been transformed to a genome browser fully devoted to regulatory genomics. Its access paradigm for genomic data is zoom-to-the-DNA-base in real time. A more recent component of the CYRENE project is the Virtual Sea Urchin system (VSU), an interactive visualization tool that provides a four-dimensional (spatial and temporal) map of the gene regulatory networks of the sea urchin embryo.
Key words: cis-regulatory architecture, gene regulatory networks, transcription factors, cisGRN- Lexicon, cisGRN-Browser, virtual sea urchin.
1. Introduction
When the GRN context is clear, we will use at times the shorthands “cis-Lexicon” and “cis-Browser”. The cis-Lexicon and the cis-Browser are conceptually two separate entities, a database
I. Ladunga (ed.), Computational Biology of Transcription Factor Binding, Methods in Molecular Biology 674, DOI 10.1007/978-1-60761-854-6_22, © Springer Science+Business Media, LLC 2010 369 370 Istrail et al.
and a visualization tool; however, from the user’s point of view, they are intertwined into one integrated software environment for regulatory genomics. The cisGRN-Lexicon is a database containing the regulatory architecture (the genomic regulatory region) of a set of transcrip- tion factor-encoding genes as well as of a number of other regula- tory genes. This architecture is presented with full genomic struc- ture known to date, including transcription factor binding site sequences, the organization into cis-regulatory modules (CRMs), and various other types of functional genomics (e.g., logic func- tions) annotations of the DNA regulatory region revealed by cis-regulatory analyses and systematic experimental perturbations of gene regulatory networks. The cisGRN-Lexicon annotations, accessible though the cisGRN-Browser, include the transcription factor binding site, the trans acting factor, the protein family to which the trans acting factor belongs, the cis-Regulatory Mod- ule (CRM) boundaries, the spatial and temporal functionality of the CRM, and the molecular function of the encoded protein. The cisGRN-Lexicon is embedded in and accessed through the cisGRN-Browser and is supported by various software libraries of tools for cisGRN-Lexicon annotators. One such system under development is CLOSE (cis-Lexicon Search Engine), a set of algorithmic strategies for literature extraction of cis-regulation articles to speed identification of new CYRENE genes and esti- mate the “dimension” of the CYRENE gene universe. We describe the current state of the CYRENE cisGRN- Browser: its detailed architecture and its planned improvement in partnership with scientists from the Davidson Lab at the Cali- fornia Institute of Technology, where the cis-Browser has been in use for the past few years (18). The Virtual Sea Urchin software system aims at giving a three- dimensional representation of the embryo’s cellular anatomy stages in which gene expression is represented in time and within specific cell types. It will be integrated with the cisGRN-Browser and BioTapestry (19, 20)topresentaView from the GRNs.
1.1. Algorithms for The object of the cis-Lexicon is to create a data set that makes pos- CRM Regulatory sible the prediction of cis-regulatory elements in DNA sequences Architecture of unknown function. To achieve a better prediction algorithm, Prediction the data set used must not be contaminated by low-quality data. Furthermore, the categories for annotation of a gene into the lexi- con must be based on a relevant model of evolution and biological function. Helpful surveys describing the state of the art of algorithms for prediction of sites, modules, and organization of regula- tory regions are refs. 21–24. Many algorithms presented in the literature have aimed at addressing various aspects of the com- putational prediction problems related to regulatory genomics Practical Computational Methods for Regulatory Genomics 371
(25–44, 19, 45–47). The present work aims at providing a database of cis-regulatory architectures, experimentally validated at the highest level as a basis for the design of the next generation of regulatory prediction algorithms.
1.2. CYRENE Genes The cisGRN-Lexicon presently contains the regulatory archi- and GRNs tecture of 200 transcription factors encoding genes and 100 other regulatory genes in eight species: human, mouse, fruit fly, sea urchin, nematode, rat, chicken, and zebrafish. The regula- tory architecture of each of these CYRENE genes contains only functionally authenticated sites by site-specific mutagenesis, con- ductedinvivo, and followed by gene transfer and functional test. As the objective is to determine how genes are regulated in vivo, it follows that only in vivo cis-regulation studies should be admitted to the cis-Lexicon. This database differs from other databases of gene regula- tion in several ways. It will be displayed in an interactive way that presents on one page the whole genome, a workspace for cis-regulatory analysis, and all the relevant gene functions. Many annotation categories and functions are unique to the cis-Lexicon. Experimental evidence required for admittance to the lexicon is stringently examined.
1.3. The Need for Combining GRN inference experiments (identification of regula- Integrated Cell tory genes (41, 48–50), perturbation experiments (51–52, 49, Models 53), cis-regulatory analysis (11, 13, 54, 55), etc.) with recent developments in systems biology imaging (56, 57) will make pos- sible the construction of a full 4D spatiotemporal map of the sea urchin embryo. Such a map will fully describe the intra- and intercellular interactions of the GRN in the developing sea urchin embryo. The analysis and visualization tools needed to interpret these data seem to have lagged far behind experiments. Thus, we have been developing a natural visualization and analysis environ- ment – the Virtual Sea Urchin (VSU) – that allows researchers to interrogate the developmental atlas at any time and at any posi- tion in the developmental process.
2. Materials
2.1. cisGRN-Browser: The CYRENE cis-Browser was developed in the Eclipse IDE Software for Java Developers (http://www.eclipse.org). The foun- dation of the cis-Browser is the Celera Genome Browser (58), whose source code is available free on Source- Forge.net (http://sourceforge.net/projects/celeragb/). The 372 Istrail et al.
cis-Lexicon is stored as an Apache Derby Database (http://db. apache.org/derby/).
2.2. Virtual Sea Our system is composed of OGRE (an open-source 3D graphics Urchin: Software engine), a C++ visualization application, and data describing the GRN such as transcription factor binding site affinity and prod- ucts, cell-signaling pathways, etc. Three-dimensional sea urchin embryonic models were created using Blender. Future directions will include the integration of the regulatory network simulator BioTapestry (19, 20)andthecisGRN-Browser Cyrene. Portions of the embryo viewing application were provided by OgreMax (http://www.ogremax.com).
3. Methods
3.1. Cis-Lexicon 3.1.1. Anatomy of the The clues given by biology for the rules of cis-regulation are copi- Lexicon ous; the difficulties lie in creating criteria to represent the clues in a meaningful way. We must thus develop a classification system that neither oversimplifies biology, so that categories lose their physical meaning, nor overcomplicates the issues by creating more categories than necessary. While biology is inherently resistant to precise definitions, categories are necessary for the sake of high- throughput data clustering. The many challenges in creating con- trolled vocabularies are discussed shortly. The vocabularies of the cis-Lexicon are based on two guiding principles: (1) vocabularies should represent phenomena in a way that fits their physical inter- action and (2) vocabularies should facilitate comparison. Where possible, the vocabularies are externally linked in such a way that when the vocabulary is updated, so is the cis-Lexicon. Cis-Lexicon annotations include the transcription factor binding site (TFBS), the function of the TFBS, the trans acting factor, the protein family to which the trans acting factor belongs, the cis-regulatory module (CRM), the spatial and temporal func- tionality of the CRM, and the molecular function of the encoded protein. These categories in the cis-Lexicon were developed for useful data clustering. When a gene could not be annotated accu- rately within the constraints of our chosen vocabulary, new cate- gories were created in order to categorize the cis-regulatory archi- tecture in a biologically relevant manner.
3.1.2. cis-Regulatory Transcription factor binding sites (TFBS) usually span 6 to 8 bp, Ontology though sometimes many more. Every TFBS in the lexicon is annotated as performing one or more of the following functions (more functions may exist in the natural world, but this is the set Practical Computational Methods for Regulatory Genomics 373
Fig. 22.1. Screenshot from the cis-Browser of the Drosophila transcription factor encoding gene, eve.
of all functions encountered so far) [See Fig. 22.1 for an example of cis-regulatory function in the cisBrowser]: • Repression – Indicates that mutating the TFBS increases gene expression or produces ectopic expression. Repressors may act “long range,” when the repression effect may tar- get more than one enhancer, or “short range,” when repres- sion affects only neighboring activators (59, 60). The func- tion of repression applies in cases where the repressors inter- act with the basal transcription apparatus either directly or indirectly (61). • Activation – Indicates that mutation decreases gene expres- sion. An activator TFBS may act over a large genomic dis- tance or short. See Latchman (62) for further discussion of some of the many ways a transcription factor can accomplish activation. • Signal response – Indicates that the transcription factor has been shown to be activated by a ligand such as a hormone (phosphorylation is not included) (63). • DNA looping – Indicates that the binding factor is involved in a protein–protein interaction with another binding factor some distance away that causes the DNA to form one or 374 Istrail et al.
more loops. This looping brings distant regulatory elements closer to the basal transcription apparatus (64). • Booster – Indicates that the TFBS does not increase gene expression on its own but can augment activation by other TFBSs. • Input into AND logic – Indicates that the TFBS can activate gene expression only when two or more cooperating TFBSs are bound. Assigned to one of at least two TFBSs (14). • Input into OR logic – Indicates that the TFBS can activate gene expression when either or both of two or more coop- erating TFBSs are bound. Assigned to one of at least two TFBSs (14). • Linker – Indicates that a TFBS is responsible for communi- cating between CRMs. • Driver – Indicates that this TFBS is the primary determining factor of gene expression. The binding factor appears only in certain developmental situations and thus is the key input for directing gene expression. TFBSs that are not drivers usually bind ubiquitous factors (65). • Communication with BTA (basal transcription apparatus). • Insulator – Indicates that the TFBS causes cis-regulatory ele- ments to be kept separate from one another. Insulators can separate the cis-regulatory elements of different genes as well as act as a barricade to keep active segments of DNA free of histones and remain active (66).
3.1.3. The trans Acting The transcription factor binding to the regulatory DNA is anno- Factor tated as the gene name given in NCBI rather than the name given to the factor in the literature. For example, while Inagaki refers to human TF c-Jun (67), this is annotated in the cis-Lexicon as JUN for consistency. Each transcription factor in the lexicon is also assigned to a leaf of a transcription factor hierarchy adapted from TRANSFAC (68)(see Figs. 22.2 and 22.3)). More closely- related transcription factors may behave more similarly, so that when the data in the cis-Lexicon are clustered, patterns may be found by grouping transcription factors according to their evolu- tionary origins.
3.1.4. cis-Regulatory TFBSs occur in groups and each grouping usually directs gene Modules expression in one temporal and spatial location. The CRM (1) includes the binding sites responsible for gene expression as well as the neighboring sequence established to enable the TBFSs to function correctly. Each CRM in the lexicon is annotated as functioning in a specific spatial and temporal location. There is currently no associated ontology for annotating this location in the lexicon. Exhaustively naming all locations and time points of Practical Computational Methods for Regulatory Genomics 375
Other Pocket Domain HMGI(Y) Other E1A-like-factors AP2/ERBP-related factors PBC XANF
Zinc-fingers of alternating composition
Diverse Cys4
Cys4 (Nuclear Receptor)
Cys2His2
TATA-binding STAT SMAD/NF-1 Runt RHR p53 MADS box HMG Histonefold Eve Grainyhead
Cold-shock domain factors Dll Ems En Prd HD-ZIP H2.0 HNF1 Msh Nk-2 Cut Tryptophan Cluster eve TEA Domain POU domain factors EVX GBX HLXB9 MEOX Paired Box Paired box plus homeodomain Homeodomain Homeodomain with LIM region Antp Cad Heat Shock Factors Homeodomain only AbdB Forkhead/Winged Helix Homeodomain plus Zinc-finger motifs
bHSH bHLH-ZIP bHLH bZIP Basic Domains Helix-turn-Helix beta-Scaffold Zinc-containing Fig. 22.2. Transcription factor hierarchy (homeodomain family expanded as an example).
an organism’s development is beyond the scope of this project, though a controlled vocabulary of body parts and stages would be useful.
3.1.5. Gene Functions Each gene whose cis-regulatory architecture is annotated in the lexicon is assigned to one of seven gene function categories. Many ontologies of gene functions already exist, such as the Gene Ontology (69) and Panther Classification System (70), but the cis-Lexicon Gene Functions ontology was created with the spe- cific intent of grouping gene functions so that genes with similar cis-regulatory architecture are grouped together. The hypothesis is that housekeeping genes have cis-regulatory architecture dis- tinct from transcription factor-encoding genes or signaling genes. GO annotations for each gene indicated in brackets show simi- larities and differences in gene function annotation between GO and the cis-Lexicon. See Fig. 22.4 for gene functions in the cis- Lexicon. 376 Istrail et al.
70
60
50
40
30
TFs in cis-Lexicon 20
10
0 p53 STAT RHR bZIP Runt HMG bHLH bHSH HMGI(Y) Cys2His2 bHLH-ZIP MADS box Paired Box Paired Grainyhead Histone fold SMAD/NF-1 TEA Domain Diverse Cys4 Homeodomain Pocket Domain E1A-like factors Tryptophan Cluster Tryptophan Heat Shock Factors Copper fist porteins TATA-binding proteins TATA-binding beta-Barrel alpha-helix beta-Barrel Fork Head/Winged Helix Cys4 (Nuclear Receptor) Cold-shock domain factors AP2/EREBP-related factors AP2/EREBP-related zinc fingers of alternating composition Fig. 22.3. Distribution of transcription factor encoding genes annotated in the cis-Lexicon categorized by transcription factor family (68).
250
200
150
100 genes in cis-Lexicon 50
0
miRNA Nuclear Signaling Cell Cycle ExtracellularCytoskeletal Housekeeping Differentiation
Transcription Factor Fig. 22.4. Distribution of gene functions for which the cis-regulatory architecture has been annotated in the cis-Lexicon. Practical Computational Methods for Regulatory Genomics 377
• Cell cycle – Genes involved in cell division that decide when to replicate DNA, when to divide the cell, etc., especially before cells terminally differentiate. Example: cyclins run in the cis-Lexicon: IL3 (Homo sapiens)(71) [extracellular space, cytokine activity, growth factor activity, etc.]; BCL2 (H. sapi- ens)(72) [apoptosis, etc.]. • Cytoskeletal – Genes involved in maintaining the structure of the cell, for example, actin and tubulin. In the cis-Lexicon: myh4 (M. musculus)(73) [actin binding, myosin filament, etc.; ASL (G. gallus)(74) [structural constituent of eye lens]. • Differentiation – Genes “expressed in the final stages of given developmental processes... They receive rather than gener- ate developmental instructions” (2). In the cis-Lexicon: mal- pha (Drosophila melanogaster)(75) [cell fate specification, notch signaling pathway, sensory organ development]. • Extracellular – Genes whose product is released from the cell, such as hormones. These do not include membrane pro- teins, which could be categorized as housekeeping or signal- ing. In the cis-Lexicon: Col5A2 (H. sapiens)(76) [eye mor- phogenesis, skin development, extracellular matrix structural constituent, etc.]; SOD3 (H. sapiens)(77) [cytoplasm, extra- cellular region, zinc ion binding, etc.]. • Housekeeping – Genes continuously expressed that regulate processes inside the cell such as transcription apparatus, ribo- somes, degradation proteins, and many enzymes. In the cis- Lexicon: btl (D. melanogaster)(78, 79) [endoderm develop- ment, glial cell migration, negative regulation of axon exten- sion, etc.]; BACE1 (Homo sapiens)(80) [proteolysis, pepti- dase activity, etc.]. • Transcription factor – Genes whose product binds to DNA in a sequence-specific manner to affect gene expression. Does not include basal transcription factors. In the cis-Lexicon: twi (D. melanogaster)(81–84), [specific RNA polymerase II transcription factor activity, etc.]; Hoxa4 (H. sapiens)(85); Foxa2 (Mus musculus)(86, 87) [RNA polymerase II tran- scription factor activity, enhancer binding, etc]. • miRNA – Genes encoding micro RNAs. In the cis-Lexicon: DmiR-1 (D. melanogaster)(88) [cardiac cell differentiation, regulation of notch signaling]. • Signaling – Genes acting as part of a signaling pathway such as hormones, hormone receptors, and kinases. In the cis-Lexicon: IL4 (M. musculus)(89) [extracellular space, B cell activation, interleukin-4 receptor binding, etc.]; ins2 (R. norvegicus)(90) [cytoplasm, extracellular space, hor- mone activity, etc.]. 378 Istrail et al.
3.1.6. Quintessential The categories used in the literature to classify the function of Diagram Problem a TFBS are often described in simple terms that prevent the annotator from fully describing the function according to the cis- regulatory ontology (see Section 3.1.2). In the literature, a TFBS is generally declared an “activator” if deletion lowers output and a “repressor” if deletion increases output. Other more complex mechanisms may cause increased or decreased expression, such as DNA looping, communication with basal transcription apparatus, etc., but these are often unreported in the literature. Ideally, these more complex mechanisms would be known for each trans act- ing factor in the cis-Lexicon. Such biochemical clues would make possible effective data clustering, thus presenting clues for pre- dicting cis-regulation. For example, DNA looping between two TFBSs cannot occur at less than a certain minimum distance, while Su(H), a transcription factor activated by signaling, may have a maximum distance from the transcription start site while still being able to direct gene expression of the sea urchin gene, gcm (9). Our lexicon is designed to handle more complex fields, but this information is not always available. Perhaps more cis-regulatory information can be derived from the quantitative data obtained by mutating a TFBS (Fig. 22.5). Most literature containing data that meet the criteria of the cis-Lexicon contains a bar chart quantitatively describing gene expression as a result of mutating each of the TFBSs individu- ally and in combination (18). Gene expression is thus a function of each of the inputs. Gene expression, the output of the function, is the combined effect of each of the individual inputs. This func- tion is not simply the sum of the effect of each individual input; rather, the output depends on the interaction of the inputs. Thus, describing the gene expression requires a more complicated func- tion than summing the effects of mutating each TFBS individ- ually. A generalized mathematical function has been suggested, but applying the function to a broad range of mutational studies is difficult (15). The cis-Lexicon currently does not handle the annotation of the quantitative data from literature. Knowing the relative impact of each TFBS on gene expression is important in properly describ- ing cis-regulatory architecture. Thus a format for collecting these data that describe the biology effectively needs to be implemented from quantitative experimental data. Since gene expression can depend on which nucleotides are mutated within a TFBS, there is a relatively low certainty associated with these quantitative data, adding further complexity to the problem.
3.1.7. Examples in the See Figs. 22.6 and 22.7. Lexicon Practical Computational Methods for Regulatory Genomics 379
myh3 in Mus musculus c ); 91 (
Gallus gallus alphaA-crystallin in b ); 9 (
Strongylocentrotus purpuratus sp-gcm in a -regulatory architecture.
cis ). 92 Fig. 22.5. Quintessential diagrams( of 380 Istrail et al.
Fig. 22.6. Comparison of four pax genes in the cis-Lexicon (axis not to scale). The purple square represents the first exon, the yellow double-ended bar represents the CRM, and the orange blocks represent the TFBS.
10000
1000
in cis-Lexicon 100 in Genome
number of genes 10
1
D. rerio* (0.2%) G. gallus* (0.4%) C. elegans‡ H.(0.7%) sapiens** M.(3.5%) musculus** (4%) R. norvegicus**S. (0.9%)purpuratus† (0.6%) D. melanogaster** (4.2%) Fig. 22.7. Annotation progress for transcription factor encoding genes. The percent in parentheses is the percentage of transcription factor encoding genes in the corresponding genome that have been annotated in the cis-Lexicon. Genome- wide transcription factor encoding gene totals were taken from the following sources: ∗ –DBD(93), ∗∗ – Panther (70), † – SpBASE (48), ‡ –(94).
3.2. cis-Browser 3.2.1. Development of The CYRENE cis-Browser is a genome browser tailored for cis- the cis-Browser regulatory annotation and investigation. It began as a branch of the Celera Genome Browser, which is available as open source at http://sourceforge.net/projects/celeragb/. The features of the original Celera Genome Browser centered on viewing and anno- tating gene transcripts, so many new capabilities were added to address our new focus. First, support for cis-regulatory modules (CRMs) and tran- scription factor binding sites (TFBSs) was added. Each of these new genomic features possesses several unique properties and Practical Computational Methods for Regulatory Genomics 381
associated information. Unlike gene transcripts, whose borders are determined solely by their exons, the boundaries of CRMs can extend beyond the known binding sites contained inside (e.g., if evidenced by sequence conservation). It is often known whether or not whole CRMs or individual binding sites are conserved across species, and this information can be added and viewed via the cis-Browser. Each TFBS has a specific factor (or, occasionally, a family of related factors) that binds there. The NCBI GeneID or name for this factor and its effect on gene expression can be annotated and viewed in the cis-Browser. Support for new types was added by creating new Java classes. For example, the class CuratedCRM was created to represent CRMs, and the existing class CuratedTranscript (from the Celera Genome Browser) was used as a reference, since transcripts and CRMs share key traits. The focus of the cis-Browser is on annotations that are sup- plemental to known genes, rather than on discovering transcripts. Therefore, instead of requiring annotators to input the genes themselves, the capability was added to download genes directly from NCBI. Within the cis-Browser application, the user can search for genes (see Fig. 22.8) just as in the NCBI Entrez Gene web site. When a gene is selected from the results, the genomic sequence of the region is downloaded and all the gene’s tran- scripts and exons are automatically displayed. Data are accessed via the NCBI Entrez Programming Utilities service (http:// eutils.ncbi.nlm.nih.gov/entrez/query/static/eutils_help.html). In the Celera Genome Browser, properties of genomic enti- ties could be of two types: plain text (e.g., names) or a choice from a list of options (e.g., evidence type: cis-mutation, footprinting, etc). Properties could be nested, so that a single (parent) property could contain inside it several additional (child) properties. For cis-regulatory annotation, we required accurate recording of com- plex properties. First, we needed to support properties containing
Fig. 22.8. Searching NCBI Entrez Gene within the cis-Browser. 382 Istrail et al.
multiple interdependent parts: for example, when annotating the factor that binds at a certain site, we must keep track of the fac- tor name, its NCBI GeneID, and any synonyms mentioned in the literature, so that they do not fall out of sync. Second, we needed to support multiple values for a single property: multiple synonyms, multiple cis-regulatory functions, and conservation in multiple species. For properties with multiple parts, we created rich dialog boxes ensuring that the user enters correct information. It would be tedious to ask the user to flip back and forth between the cis- Browser and the NCBI Entrez Gene web site to find GeneIDs for each binding factor, and it would be error-prone to make the user type in the factor names and GeneIDs, especially when the same factor binds at several sites for a single target gene. Therefore, the cis-Browser provides a special window for annotating the factor that binds to each site (see Fig. 22.9), so that the user can search the Entrez Gene site from within the browser; the search is auto- matically restricted to the species being annotated. If the same factor binds at multiple sites, for the second and later sites the user can select the gene from a menu rather than re-entering the information. There are similar windows for annotating conserved species (which searches NCBI for the correct scientific names and NCBI ID) and cis-regulatory functions (which ensures that the cis-regulatory ontology is followed in naming the regulatory
Fig. 22.9. Bound factor annotation with search results. Practical Computational Methods for Regulatory Genomics 383
Fig. 22.10. Conserved species annotation with search results. functions and verifies that PMIDs are typed correctly) – see Figs. 22.10 and 22.11. Three different mechanisms were tested to support pro- perties with multiple values; the first two are described in Notes section below. The problem was that in the Cel- era Genome Browser data model, only one property value can be assigned to a given name. In the GAME XML format supported by the browser, this is represented by the tag
Fig. 22.11. Regulatory function annotation.
At one point we created an ad hoc index as an interim measure to allow certain restricted queries (e.g., list all genes in the lexicon or ask whether a given gene is in the lexicon). Later, we imple- mented the cis-Lexicon as a true database using Apache Derby, an open-source relational database engine. Since Apache Derby is implemented entirely in Java, the cis-Browser remains entirely cross-platform. Derby can be run either in embedded mode, where the database is stored and accessed locally, or as a network client, where the database is stored remotely and accessed via a server. This allows the cis-Lexicon to be packaged with the cis- Browser, for ease of access, or to be stored in one central loca- tion, for ease of updating. Relational databases such as Derby support automatic indexing of specific fields in a database record. By indexing the NCBI GeneID field of gene records, for exam- ple, searching for genes by GeneID immediately becomes fast. Relational databases also support foreign keys, which ensure that Practical Computational Methods for Regulatory Genomics 385
values intended to reference other entities are actually valid. For example, if the bound factor for a binding site is recorded as gene 373400, then the database either ensures that gene 373400 is present in the cis-Lexicon or rejects the annotation.
3.2.2. The CYRENE The cis-Browser interface has the same organization as the orig- cis-Browser Interface inal Celera Genome Browser. The cis-Browser application win- dow is split into four regions (clockwise from top left): the Out- line View, the Annotation View, the Subview Container, and the Property Inspector View (see Fig. 22.12). The Outline View dis- plays in a hierarchical tree format the species, chromosomes, and sequences loaded by the cis-Browser and ready for analysis. The Annotation View displays the locations of genomic features (e.g., transcripts, CRMs) on the sequence currently being examined. The Subview Container shows the user a set of views specific to the currently selected feature, and the Property Inspector View shows the properties of the selected feature in textual form. The Annotation View allows real-time zooming from a chromosome-wide view down to the individual nucleotide level. When the user clicks on a genomic feature, information specific to the feature is visible in the Subview Container and the Property Inspector View. The Annotation View displays genomic features in tiers (horizontal rows grouping features according to their source) so that information from multiple sources is not inter- mixed and confused; in Fig. 22.13, for example, mapped Solexa
Fig. 22.12. The cis-Browser window. 386 Istrail et al. Fig. 22.13. The gene brn1/2/4 appears in the “Promoted” tier while mapped Solexa reads appear in the “Solexa” tier. Practical Computational Methods for Regulatory Genomics 387 reads are grouped separately from genomic transcripts. One spe- cial tier in the Annotation View is the workspace, which is the tier that contains genomic features currently being edited. Features loaded from data sources such as XML files or the cis-Lexicon are considered immutable, so a copy must be made and placed in the workspace before it can be modified. New features added by the annotator, such as CRMs and binding sites, are always in the workspace. Each type of genomic feature has particular traits that dis- tinguish it from other types. For example, transcripts are trans- lated into proteins and BLAST hits are the result of comparisons between different sequences. Therefore, viewing the translation of codons into amino acids is relevant only for transcripts, while examining the differences between the current sequence and a sequence that was searched against it is relevant only for BLAST hits. The Subview Container is the location for views such as these. When a feature is selected in the Annotation View, only the views relevant to that type of feature are shown in the Subview Container. Only one such view is shown at a time, to maximize the visible area; the rest are shown as tabs that the user may click on to switch to that view. Every genomic feature has certain associated properties, such as name, NCBI accession number, or date of curation. The Prop- erty Inspector View displays these properties as a two-column table giving the name of each property on the left and the value on the right. Properties can be edited by double-clicking the cur- rent value. If the value is a simple string, then it can be edited in place; if it is more complex, such as binding factors, then a dialog box appears. Property changes affecting how the feature appears in the Annotation View are reflected in real time: when the name of a gene or CRM is modified, the new name appears immediately. One subview (i.e., view appearing in the Subview Container) of critical importance is the Consensus Sequence View, which dis- plays the sequence of the selected feature and the surrounding region and is also used to mark the location of new features. The user simply clicks and drags to select a sequence. Right-clicking shows a menu with options to create a transcript, CRM, or TFBS. The seqFinder feature quickly locates the exact coordinates of a sequence appearing in a published paper. Given a region of sequence to search within (e.g., a gene and its flanking sequence), the seqFinder lets the user type in only the minimum number of nucleotides to uniquely find the paper’s sequence. For each letter the user types, the seqFinder tells whether the sequence typed so far is found more than once (i.e., multiple ambiguous matches, so more input is necessary), exactly once (i.e., a perfect match; no more typing needed), or never (i.e., a typo or possibly a true mis- match between the paper’s sequence and the reference genome). 388 Istrail et al.
The user need only type a few letters from the beginning and then from the end to uniquely identify the entire sequence. When the start and end coordinates are known, the seqFinder automatically selects the sequence within the Consensus Sequence View. Typing the minimum sequence necessary lets the user locate the precise coordinates quickly yet accurately.
3.2.3. Annotation with To enter a genomic feature, the annotators first input the coordi- the cis-Browser nates by locating them with the seqFinder. The relevant proper- ties such as names, binding factors, cis-regulatory functions, and sequence conservation are set via the Property Inspector View. The Annotation View lets one do quick sanity checks – are the binding sites located upstream, downstream, or within introns of the regulated gene, as is usually the case? Are the CRMs of a rea- sonable size? The annotators’ work is saved as XML files in the GAME for- mat, rather than directly input into the cis-Lexicon. This allows easy backup and sharing of past work and also prevents clutter- ing the database with half-finished or faulty annotations. A spe- cial software tool is required to move the annotations from these intermediate files into the cis-Lexicon; forcing the use of inter- mediate files and preventing unauthorized annotators from mod- ifying the cis-Lexicon directly lets us keep the database at a strict high quality. An experienced annotator can verify the work of a trainee before it is entered into the cis-Lexicon.
3.3. Virtual Sea We have worked closely with the Davidson laboratory at CalTech Urchin to produce a Virtual Sea Urchin prototype. The VSU uses spatial models and a graphics engine to simulate the four-dimensional sea urchin embryo, allowing the researcher to probe the GRN at levels of granularity from the multicellular embryo to the gene- regulatory network of an individual cell type. The embryo models were created by extrapolating to three dimensions cross-sectional color-coded tracings from photomicrographs (17). The Virtual Sea Urchin currently provides models for the Strongylocentrotus purpuratus embryo at 6, 10, 15, 20, and 24 h. Cell types are defined by ambient and diffuse coloring as well as shape. Gene expression data are visible at a glance on an embry- onic cell type using emission coloring (intensity of coloring is pro- portional to intensity of expression). The VSU model of embryonic development will eventually be configurable, featuring realistic cell models and dynamics sim- ulators. In toto imaging (56, 57)oftheseaurchinembryowill enhance the model’s accuracy and resolution, letting researchers probe the regulatory network activity per cell. We will ulti- mately combine the cis-regulatory sequence-analysis capabilities of CYRENE and the network building, visualization, and simu- lation capabilities of BioTapestry with the temporal and spatial Practical Computational Methods for Regulatory Genomics 389
Fig. 22.14. Virtual Sea Urchin, BioTapestry, and Cyrene integration. In order to model the gene regulatory network of the developing sea urchin embryo completely, tools that specialize in analyzing different perspectives of the network must be integrated. BioTapestry operates at the network level, allowing users to manipulate and simulate network interactions. Cyrene operates at the DNA level, providing cis-regulatory module and binding site definition and other DNA sequence analysis. The Virtual Sea Urchin maps regulatory network information into space and time. The interoperability among these three views is a central component of future work.
analysis of the 4D Virtual Sea Urchin to yield a complete charac- terization of the GRN (see Fig. 22.14).
4. Notes
1. Gene naming Creating a controlled vocabulary starts with the name of the gene, as discussed previously (18), although there has been a large effort to establish common names across species (95). We would like to know the primary ortholog of some gene we have annotated to observe the similarities of cis- regulatory architecture across species. Our efforts toward 390 Istrail et al.
Table 22.1 Examples for gene name translation between mouse and Drosophila Mouse Drosophila References
Beta-catenin Armadillo van Noort et al. (2002) (103) Cux-1 Cut Sharma et al. (2004) (104) TLE-4 Groucho Sharma et al. (2004) (104) six3, six6 So López-Ríos et al. (2003) (105)
such a gene name translation table are only in the nascent stages. Examples for gene name translation between mouse and Drosophila are shown in Table 22.1. A gene occurring in two species is more likely to have a similar cis-regulatory architecture if the two genes are related by evolution and have a conserved function. Sets of genes meeting these cri- teria we have termed Davidson Orthologs. The method usu- ally employed for determining homology is by sequence rather than conserved function. orthoMCL (96)andinpara- noid (97) are examples of this kind of ortholog table. While the latter definition of ortholog is more easily searched in a database of genes automatically, the definition is not as strin- gent. 2. CRM boundaries TFBSs are usually well described and require no guesswork by the annotator, but CRMs are often not well defined in the literature. Some of this confusion stems from the lack of a precise definition of the function of a CRM; addition- ally, many research groups are not interested in finding the boundaries of CRMs and limit their scope to discovery of TFBSs. When annotating the cis-regulatory architecture of a gene, the annotator often must make certain assumptions about the boundaries of a CRM that can be classified as follows (examples are referenced): (1) CRMs are not dis- cussed in the literature and the annotator defines the CRM by the minimal sequence that correctly directs gene expres- sion, usually approximated to within 100 bp (96). (2)CRMs are not discussed in the literature, but a graph in the paper shows sequence similarity to the same gene in other species (98). The annotator defines the CRM as the sequence most conserved in other species. If a sequence remains highly sim- ilar over a great evolutionary distance, there must be selec- tive pressure to conserve the sequence, and therefore the sequence probably plays an important role in the organ- ism. (3) CRMs are not discussed in the literature and the Practical Computational Methods for Regulatory Genomics 391
annotator defines the CRM by most extreme TFBSs deter- mined to act in a specific location (99). That is, if three TFBSs are found to drive gene expression in a cell, the CRM annotation goes from the first nucleotide of the first TFBS to the last nucleotide of the last TFBS. Overall uncertainty and lack of consistency in CRM annotation reduce the quality of data on CRM boundaries. 3. Caveats in Davidson criteria While the cis-Lexicon seeks to collect only the most reliable data, many uncertainties remain. Mutational studies show that a certain TFBS is important for correct gene expression, but the factor that binds to the site is not immediately cer- tain. The sequence probably contains a transcription factor binding motif, so the factor that binds can often be guessed. In some annotations, the experimentalist has shown that a particular transcription factor binds to the TFBS in an assay (100) or that knockdown of the transcription factor also causes a change in gene expression. Such confirmations of trans acting factor are not always reported in the literature. 4. Uncertainty in identifying the trans acting factor Often in the literature the exact transcription factor bind- ing to a TFBS is not known, but the transcription factor family to which the trans acting factor belongs is reported; Shen and Ingraham (101) report an E-Box trans activator. Sometimes the authors report that the trans acting factor could be one of several; for instance, Clark et al. (102) report that either an RAR/RAR or RAR/RXR dimer activates tran- scription. The TFBS cannot be compared to others in the cis-Lexicon since the trans acting factor is not known, and the TFBS loses its usefulness in the data set for clustering and prediction. The transcription factor hierarchy described above (Fig. 22.2) was added as an annotation tool to com- bat this problem. Trans acting factors can be clustered at different levels of the transcription factor hierarchy. 5. cis-Lexicon search engine A great challenge in the annotation process has been find- ing literature relevant to building the cis-Lexicon. So far the literature has been located by PubMed or Google Scholar searches or by browsing references describing previously annotated genes (Fig. 22.15 shows journals cited in the cis- Lexicon). A formalized search process will rapidly uncover relevant literature; in addition, it will help determine the number of genes studied according to the Davidson crite- ria and give an estimate of cis-Lexicon completeness. When the cis-Lexicon is declared complete, searches will have to be performed continually to find new data. To accomplish these goals, the CLOSE Project (cis-Lexicon Ontology Search 392 Istrail et al.
60
50
40
30
20 number of citations
10
0
Cell Gene PNAS Blood Other Oncogene Development
The EMBO Journal Genes & Development Developmental Biology Nucleic Acids Research The BiochemicalMolecular Journal EndocrinologyDevelopmental Dynamics Journal of Molecular Biology Molecular and Cellular Biology
The Journal of Biological Chemistry
Journal of Molecular and Cellular Cardiology
Biochemical and Biophysical Research Communications Fig. 22.15. Journals referenced in the cis-Lexicon.
Engine) aims to create a set of algorithmic strategies for lit- erature extraction of cis-regulation articles. PubMed cannot perform unrestricted phrase search- ing of citations and abstracts; only phrases in the PubMed Index are found. If a phrase is not in the Index, then a PubMed search cannot return exact matches, even if it appears in citations or abstracts. Instead, the query is treated as a standard non-phrase search, yielding almost entirely irrelevant results (http:// www.nlm.nih.gov/bsd/disted/pubmedtutorial/020_450. html). This prevents the use of PubMed for advanced text-based queries. MeSH terms are not assigned consis- tently enough to make possible comprehensive searches by keyword alone (see, for example, Yuh et al. (13), which is not assigned a term for Base Sequence). Google does not yet offer an API to access Google Scholar, and the results of searching PubMed via standard search engines (Google, Yahoo, etc.) lack known important papers (unpublished Practical Computational Methods for Regulatory Genomics 393
data). All of these point to the need for a specialized search engine to find the most relevant papers. 6. Complex annotations in the cis-Browser Three different mechanisms for making complex annota- tions with multiple values were tested. The first was to extend the GAME XML format with a new tag for each new type of data. This method was explored first because supporting multiple complex values appeared analogous to storing annotator comments, which was already sup- ported by the Celera Genome Browser. According to the Genome Browser data model, comments are completely dis- tinct from properties. In the XML format, each comment wasstoredina
from the NCBI Entrez Programming Utilities service always refer to DTDs located on NCBI’s servers. The XML lan- guage requires parsers always to access the DTD, and the extra time required to download the DTD whenever such an XML file is parsed led to difficult-to-trace slowdowns in the cis-Browser application. The correct solution was a custom org.xml.sax.EntityResolver implementation return- ing cached copies of NCBI files. Once given to an XML- Reader object, it will utilize the cached copies rather than fetching the remote originals. Second, computers with an old version of Java 6 may run out of memory when parsing large XML files, even when using efficient parsing methods (see http://bugs.sun.com/bugdatabase/view_bug.do?bug_id= 6536111). The easiest solutions are to use Java 5 if available or to upgrade to a more recent version of Java 6. 8. Virtual Sea Urchin embryo modeling Producing the three-dimensional models for the Virtual Sea Urchin’s 3D graphics engine is currently laborious; an embryology domain expert works with a 3D computer graphics and modeling software expert to create the embry- onic models using a software suite such as Maya or Blender. These models are then exported and integrated into a format compatible with the VSU. We can streamline this process by animating the anatom- ical structure using a hierarchy of cells and cell types that completely categorize the embryo at all relevant time slices. This hierarchical tree representation, in which each tree level defines the embryo at a specific time and which is defined by the experimentalist, can be visualized by meiotically split- ting cells into appropriate cell types. With this new embryo representation, experimentalists can interact directly with the VSU and easily define embryo development without need- ing outside expertise.
Acknowledgments
The support of the National Science Foundation under grant DBI 0645955 is acknowledged with gratitude. We would also like to acknowledge the tremendous impact on this work of our collabo- rator, Eric H. Davidson of the California Institute of Technology, who has guided every step of our efforts. This work would not have been possible without the contributions of three generations Practical Computational Methods for Regulatory Genomics 395
of annotators, most notably Tim Johnstone, Jake Halpert, and David Moskowitz. (The first generation was David Moskowitz, Rohan Madamsetti, and Sanjay Trehan; the second generation was Tamar Melman, Mark Grabiner, and Kyle Schutter; the third generation is Tim Johnstone, Jake Halpert, Mei Cao, Kenneth Estrellas, Nicole Noronha, and Daniel Yang.) We would also like to thank Andy Ransick, Andy Cameron and Russell Turner for many discussions and valuable suggestions. Last but not least, many thanks go to Erin Klopfenstein for her outstanding work and many valuable contributions to the CYRENE Project.
References
1. Davidson, E.H. (2001) Genomic regulatory ule: a key node in the sea urchin endomeso- systems: In Devel and evol, Academic Press, derm gene regulatory network. Dev Biol 269, San Diego, CA. 536–551. 2. Davidson, E.H., and Erwin, D. (2006) Gene 14. Istrail, S., De-Leon, S.-T., and Davidson, E. regulatory networks and the evolution of ani- (2007) The regulatory genome and the com- mal body plans. Science 311, 796–800. puter. Dev Biol 310, 187–195. 3. Davidson, E.H. (1968) Gene activity in early 15. Istrail, S., and Davidson, E. (2005) Logic development. Academic Press, New York, NY. functions of the genomic cis-regulatory 4. Sea Urchin Genome Consortium. (2006) code 2005. Proc Natl Acad Sci USA 102, The genome of the sea urchin Strongy- 4954–4959. locentrotus purpuratus. Science 314, 16. Levine, M., and Davidson, E.H. (2005) 941–952. Gene regulatory networks for develop- 5. Samanta, M.P., Tongprasit, W., Istrail, S. ment. Proc Natl Acad Sci USA 102, et al. (2006) The transcriptome of the sea 4936–4942. urchin embryo. Science 314, 960–962. 17. Davidson, E.H. (2006) The regulatory 6. Erwin, D.H., and Davidson, E.H. (2009) genome: gene regulatory networks in develop- The evolution of hierarchical gene regulatory ment and. Academic Press, San Diego, CA. networks. Nature Rev Gen 10, 141–148. 18. Tarpine, R., and Istrail, S. (2009) On the 7. Davidson, E.H., Rast, J.P., Oliveri, P. et al. concept of Cis-regulatory information: from (2002) A genomic regulatory network for sequence motifs to logic functions. Algo- development. Science 295, 1669–1678. rithmic Bioprocesses In (Condon, A., Harel, 8. Britten, R.J., and Davidson, E.H. (1969) D., Kok, J.N., Salomaa, A., and Winfree, E. Gene regulation for higher cells: a theory. Sci- Eds.) pp. 731–742 Springer-Verlag, Berlin ence 165, 349–357. Heidelberg. 9. Ransick, A., and Davidson, E. (2006) cis- 19. Longabaugh, W.J.R., Davidson, E.H., and regulatory processing of Notch signaling Bolouri, H. (2005) Computational represen- input to the sea urchin glial cells missing gene tation of developmental genetic regulatory during mesoderm specification. Dev Biol 297, networks. Dev Biol 283, 1–16. 587–602. 20. Longabaugh, W.J.R., Davidson, E.H., and 10. Oliveri, P., Tu, Q., and Davidson, E.H. Bolouri, H. (2009) Visualization, documen- (2008) Global regulatory logic for specifica- tation, analysis, and communication of large- tion of an embryonic cell lineage. Proc Natl scale gene regulatory networks. Biochem Bio- Acad Sci USA 105, 5955–5962. phys Acta 1789, 363–374. 11. Yuh, C.H., and Davidson, E.H. (1996) 21. Stormo, G.D. (2000) DNA binding sites: Modular cis-regulatory organization of representation and discovery. Bioinformatics Endo16, a gut-specific gene of the sea urchin 16, 16–23. embryo. Development, 122, 1069–1082. 22. Wasserman, W.W., and Sandelin, A. (2004) 12. Yuh, C.H., Bolouri, H., and Davidson, E.H. Applied bioinformatics for the identification (1998) Genomic cis-regulatory logic: exper- of regulatory elements. Nature Rev Gen 5, imental and computational analysis of a sea 276–287. urchin gene. Science 279, 1896–1902. 23. Sandelin, A. (2004) In silico prediction of cis- 13. Yuh, C.H., Dorman, E.R., Howard, M.L. regulatory elements. Karolinska Institutet. et al. (2004) An otx cis-regulatory mod- Stockholm, Sweden, 4–130. 396 Istrail et al.
24. Tompa, M., Li, N., Bailey, T.L. et al. (2005) sis in the Twilight Zone. Bioinformatics 22, Assessing computational tools for the dis- e393–e401. covery of transcription factor binding sites. 38. Badis, G., Berger, M., Philippakis, A. et al. Nature Biotechnol 23, 137–144. (2009) Diversity and complexity in DNA 25. Hannenhalli, S., and Levy, S. (2001) Pro- recognition by transcription factors. Science moter prediction in the human genome. 324, 1720–1723. Bioinformatics 17, S90–S96. 39. Berger, M., Badis, G., Gehrke, A. et al. 26. Hannenhalli, S., and Levy, S. (2002) Predict- (2008) Variation in homeodomain DNA ing transcription factor synergism. Nucleic binding revealed by high-resolution anal- Acids Res 30, 1–8. ysis of sequence preferences. Cell 133, 27. Hannenhalli, S., Putt, M.E., Gilmore, J.M. 1266–1276. et al. (2006) Transcriptional genomics asso- 40. Noyes, M.B., Christensen, R.G., ciates FOX transcription factors with human Wakabayashi, A. et al. (2008) Analysis of heart failure. Circulation J Am Heart Assoc homeodomain specificities allows the family- 114, 1269–1276. wide prediction of preferred recognition 28. Singh, L.N., Wang, L.S., and Hannenhalli, sites. Cell 133, 1277–1289. S. (2007) TREMOR—a tool for retriev- 41. Cameron, R.A., Rast, J.P., and Brown, C.T. ing transcriptional modules by incorporating (2004) Genomic resources for the study of motif covariance. Nucleic Acids Res 35, sea urchin development. Methods Cell Biol 7360–7371. 74, 733–757. 29. Markstein, M., Markstein, P., Markstein, V. 42. He, X., Ling, X., and Sinha, S. (2009) Align- et al. (2002) Genome-wide analysis of clus- ment and prediction of cis-regulatory mod- tered dorsal binding sites identifies putative ules based on a probabilistic model of evolu- target genes in the Drosophila embryo. Dev tion. PLoS Comput Biol 5, e100299. Biol 99, 763–768. 43. Li, N., and Tompa, M. (2006) Analysis of 30. Linhart, C., Halperin, Y., and Shamir, R. computational approaches for motif discov- (2008) Transcription factor and microRNA ery. Algorithms Mol Biol 1, 1–8. motif discovery: the Amadeus platform and 44. Li, X., Zhong, S., and Wong, W.H. (2005) a compendium of metazoan target sets. Reliable prediction of transcription fac- Genome Res 18, 1180–1189. tor binding sites by phylogenetic verifi- 31. Tompa, M. (1999) An exact method for cation. Proc Natl Acad Sci USA 102, finding short motifs in sequences, with appli- 16945–16950. cation to the ribosome binding site problem. 45. Papatsenko, D., and Levine, M. (2005) 7th International Conference Intelligent Sys- Quantitative analysis of binding motifs medi- tems for Molecular Biology, 262–271. ating diverse spatial readouts of the dorsal 32. Sinha, S., and Tompa, M. (2003) Perfor- gradient in the Drosophila embryo. Proc Natl mance comparison of algorithms for finding Acad Sci USA 102, 4966–4971. transcription factor binding sites. Proceedings 46. Pilpel, Y., Sudarsana, P., and Church, G.M. of the 3rd IEEE Symposium on Bioinformatics (2001) Identifying regulatory networks by and Bioengineering, 213. combinatorial analysis of promoter elements. 33. Blanchette, M., Schwikowski, B., and Nature Genet 29, 153–159. Tompa, M. (2002) Algorithms for phy- 47. Zhu, Z., Pilpel, Y., and Churge, G.M. (2002) logenetic footprinting. JComputBiol9, Computational identification of transcription 211–223. factor binding sites via a transcription-factor- 34. Wasserman, W.W., and Fickett, J.W. (1998) centric clustering (TFCC) algorithm. JMol Identification of regulatory regions which Biol 318, 71–81. confer muscle-specific gene expression. JMol 48. Howard-Ashby, M., Materna, S.C., Brown, Biol 278, 167–181. C.T. et al. (2006) Identification and charac- 35. Benos, P.V., Bulyk, M.L., and Stormo, G.D. terization of homeobox transcription factor (2002) Additivity in protein-DNA interac- genes in Strongylocentrotus purpuratus,and tions: how good an approximation is it? their expression in embryonic development. Nucleic Acids Res 30, 4442–4451. Dev Biol 300,74–89. 36. Keich, U., and Pevzner, P.A. (2002) Sub- 49. Oliveri, P., Carrick, D.M., and Davidson, tle motifs: defining the limits of motif E.H. (2002) A regulatory gene network that finding algorithms. Bioinformatics 18, directs micromere specification in the sea 1382–1390. urchin embryo. Dev Biol 246, 209–228. 37. Ng, P., Nagarajan, N., Jones, N. et al. (2006) 50. Calestani, C., Rast, J.P., and Davidson, E.H. Apples to apples: improving the performance (2003) Isolation of pigment cell specific motif finders and their significance analy- genes in the sea urchin embryo by differen- Practical Computational Methods for Regulatory Genomics 397
tial macroarray screening. Development 130, tor of sea urchin embryos capable of loop- 4587–4596. ing DNA. Proc Natl Acad Sci USA 92, 51. Imai, K.S., Levine, M., Satoh, N. et al. 2989–2993. (2006) Regulatory blueprint for a chordate 65. Smith, J., and Davidson, E. (2008) A new embryo. Science 312, 1183–1187. method, using cis-regulatory control, for 52. Ransick, A., Rast, J.P., Minokawa, T. blocking embryonic gene expression. Dev et al. (2002) New early zygotic regulators Biol 318, 360–365. expressed in endomesoderm of sea urchin 66. West, A., Gaszner, M., and Felsenfeld, G. embryos discovered by differential array (2002) Insulators: many functions, many hybridization. Dev Biol 246, 132–147. mechanisms. Genes Dev 16, 271–288. 53. Stathopoulos, A., Van Drenth, M., Erives, 67. Inagaki, N., Maekawa, T., Sudo, T. et al. A. et al. (2002) Whole-genome analysis of (1992) c-Jun represses the human insulin dorsal-ventral patterning in the Drosophila promoter activity that depends on multiple embryo. Cell 111, 687–701. cAMP response elements. Proc Natl Acad Sci 54. Revilla-i-Domingo, R., Minokawa, T., and 89, 1045–1049. Davidson, E.H. (2004) R11: a cis-regulatory 68. Matys, V., Kel-Margoulis, O., Fricke, E. node of the sea urchin embryo gene network et al. (2006) TRANSFAC and its module that controls early expression of SpDelta in TRANSCompel: transcriptional gene regu- micromeres. Dev Biol 274, 438–451. lation in eukaryotes. Nucleic Acids Res 34, 55. Lickert, H., and Kemler, R. (2002) Func- D108–D110. tional analysis of cis-regulatory elements con- 69. Ashburner, M., Ball, C., Blake, J. et al. trolling initiation and maintenance of early (2000) Gene ontology: tool for the unifica- Cdx1 gene expression in the mouse. Dev Dyn tion of biology. The Gene Ontology Consor- 225, 216–220. tium. Nat Genet 25, 25–29. 56. Megason, S., and Fraser, S. (2003) Digitizing 70. Mi, H., Guo, N., Kejariwal, A. et al. (2007) life at the level of the cell: high-performance PANTHER version 6: protein sequence and laser-scanning microscopy and image analy- function evolution data with expanded rep- sis for in toto imaging of development. Mech resentation of biological pathways. Nucleic Dev 120, 1407–1420. Acids Res 35, D247–D252. 57. Megason, S., and Fraser, S. (2007) Imaging 71. Gottschalk, L., Giannola, D., and in systems biology. Cell 130, 784–795. Emerson, S. (1993) Molecular regulation 58. Turner, R., Chaturvedi, K., Edwards, N. of the human IL-3 gene: inducible T cell- et al. (2001) Visualization challenges restricted expression requires intact AP-1 and for a new cyberpharmaceutical computing Elf-1 nuclear protein binding sites. J Exp Med paradigm, Proceedings of the Symposium on 178, 1681–1692. Large-Data Visualization and Graphics,San 72. Regl, G., Kasper, M., Schnidar, H. et al. Diego, CA. (2004) Activation of the BCL2 promoter 59. Gray, S., Szymanski, P., and Levine, M. in response to Hedgehog/GLI signal trans- (1994) Short-range repression permits mul- duction is predominantly mediated by GLI2. tiple enhancers to function autonomously Cancer Res 64, 7724–7731. within a complex promoter. Genes Dev 8(15), 73. Wheeler, M., Snyder, E., Patterson, M. et al. 1829–1838. (1999) An E-box within the MHC IIB gene 60. Courey, A., and Jia, S. (2001) Transcriptional is bound by MyoD and is required for gene repression: the long and the short of it. Genes expression in fast muscle. Am J Physiol 276, Dev 15, 2786–2796. C1069–C1078. 61. Nakao, T., and Ishizawa, A. (1994) Develop- 74. Sekido, R., Murai, K., Funahashi, J. et al. ment of the spinal nerves in the mouse with (1994) The delta-crystallin enhancer-binding special reference to innervation of the axial protein delta EF1 is a repressor of E2-box- musculature. Anat Embryol 189, 115–138. mediated gene activation. Mol Cell Bio 14, 62. Latchman, D. (2008) Eukaryotic transcrip- 5692–5700. tion factors. Fifth Edition, Academic Press, 75. Castro, B., Barolo, S., Bailey, A. et al. London. (2005) Lateral inhibition in proneural clus- 63. Barolo, S., and Posakony, J. (2002) Three ters: cis-regulatory logic and default repres- habits of highly effective signaling path- sion by suppressor of hairless. Development ways: principles of transcriptional control by 132, 3333–3344. developmental cell signaling. Genes Dev 16 , 76. Penkov, D., Tanaka, S., Di Rocco, G. et al. 1167–1181. (2000) Cooperative interactions between 64. Zeller, R., Griffith, J., Moore, J. et al. PBX, PREP, and HOX proteins modu- (1995) A multimerizing transcription fac- late the activity of the alpha 2(V) colla- 398 Istrail et al.
gen (COL5A2) promoter. J Biol Chem 275, cles during larval growth. Genes Dev 19, 16681–16689. 2343–2354. 77. Zelko, I., Mueller, M., and Folz, R. (2008) 89. Ho, I., Hodge, M., Rooney, J. et al. (1996) Transcription factors sp1 and sp3 regulate The proto-oncogene c-maf is responsible expression of human extracellular superoxide for tissue-specific expression of interleukin-4. dismutase in lung fibroblasts. Am J Respir Cell 85, 973–983. Cell Mol Biol 39, 243–251. 90. Kajihara, M., Sone, H., Amemiya, M. 78. Murphy, A., Lee, T., Andrews, C. et al. et al. (2003) Mouse MafA, homologue of (1995) The breathless FGF receptor zebrafish somite Maf 1, contributes to the homolog, a downstream target of Drosophila specific transcriptional activity through the C/EBP in the developmental control of cell insulin promoter. Biochem Biophys Res Com- migration. Development 121, 2255–2263. mun 312, 831–842. 79. Ohshiro, T., and Saigo, K. (1997) Tran- 91. Matsuo, I., and Yasuda, K. (1992) The coop- scriptional regulation of breathless FGF erative interaction between two motifs of receptor gene by binding of TRACHEA- an enhancer element of the chicken alpha LESS/dARNT heterodimers to three central A-crystallin gene, alpha CE1 and alpha CE2, midline elements in Drosophila developing confers lens-specific expression. Nucleic Acids trachea. Development 124, 3975–3986. Res 20, 3701–3712. 80. Christensen, M., Zhou, W., Qing, H. 92. Belkin, D., Allen, D., and Leinwand, L. et al. (2004). Transcriptional regulation of (2006) MyoD, Myf5, and the calcineurin BACE1, the beta-amyloid precursor protein pathway activate the developmental myosin beta-secretase, by Sp1. MolCellBiol24, heavy chain genes. Dev Biol 294, 541–553. 865–874. 93. Wilson, D., Charoensawan, V., Kummerfeld, 81. Pan, D., Huang, J., and Courey, A. (1991) S., et al. (2008) DBD—taxonomically broad Functional analysis of the Drosophila twist transcription factor predictions: new con- promoter reveals a dorsal-binding ventral tent and functionality. Nucleic Acids Res 36, activator region. Genes Dev 5, 1892–1901. D88–D92. 82. Thisse, C., Perrin-Schmitt, F., Stoetzel, C. 94. Haerty, W., Artieri, C., Khezri, N. et al. et al. (1991) Sequence-specific transactiva- (2008) Comparative analysis of function and tion of the Drosophila twist gene by the dorsal interaction of transcription factors in nema- gene product. Cell 65, 1191–1201. todes: extensive conservation of orthology 83. Jiang, J., Kosman, D., Ip, Y. et al. coupled to rapid sequence evolution. BMC (1991) The dorsal morphogen gradient reg- Genomics 9, 399. ulates the mesoderm determinant twist in 95. Bult, C., Eppig, J., Kadin, J. et al. (2008) early Drosophila embryos. Genes Dev 5, The mouse genome database (MGD): mouse 1881–1891. biology and model systems. Nucleic Acids Res 84. Akimaru, H., Hou, D., and Ishii, S. (1997) 36, D724–D728. Drosophila CBP is required for dorsal- 96. Chen, F., Mackey, A., Stoeckert, C. et al. dependent twist gene expression. Nature (2006) OrthoMCL-DB: querying a compre- Genet 17, 211–214. hensive multi-species collection of ortholog 85. Doerksen, L., Bhattacharya, A., Kannan, P. groups. Nucleic Acids Res 34, D363–D368. et al. (1996) Functional interaction between 97. Berglund, A.-C., Sjölund, E., Ostlund, G. a RARE and an AP-2 binding site in the regu- et al. (2008) InParanoid 6: eukaryotic lation of the human HOX A4 gene promoter. ortholog clusters with inparalogs. Nucleic Nucleic Acids Res 24, 2849–2856. Acids Res 36(database issue), D263–D266. 86. Sasaki, H., Hui, C., Nakafuku, M. et al. 98. Delporte, F., Pasque, V., Devos, N. et al. (1997) A binding site for Gli proteins is (2008) Expression of zebrafish pax6b in pan- essential for HNF-3beta floor plate enhancer creas is regulated by two enhancers contain- activity in transgenics and can respond to Shh ing highly conserved cis-elements bound by in vitro. Development 124, 1313–1322. PDX1, PBX and PREP factors. BMC Dev Biol 87. Yoon, J., Kita, Y., Frank, D. et al. (2002) 8, 53. Gene expression profiling leads to identifi- 99. Warren, D., Simpkins, C., Cooper, M. et al. cation of GLI1-binding elements in target (2005) Modulating alloimmune responses genes and a role for multiple downstream with plasmapheresis and IVIG. Curr Drug pathways in GLI1-induced cell transforma- Targets Cardiovasc Haematol Disord 5, tion. J Biol Chem 277, 5548–5555. 215–222. 88. Sokol, N., and Ambros, V. (2005) Mesoder- 100. Annicotte, J.-S., Fayard, E., Swift, G. mally expressed Drosophila microRNA-1 is et al. (2003) Pancreatic-duodenal home- regulated by Twist and is required in mus- obox 1 regulates expression of liver recep- Practical Computational Methods for Regulatory Genomics 399
tor homolog 1 during pancreas development. 103. van Noort, M., van de Wetering, M., and MolCellBiol23, 6713–6724. Clevers, H. (2002) Identification of two 101. Shen, J.-C., and Ingraham, H. (2002) novel regulated serines in the N terminus of Regulation of the orphan nuclear receptor beta-catenin. Exp Cell Res 276, 264–72. steroidogenic factor 1 by Sox proteins. Mol 104. Sharma, M., Fopma, A., Brantley, et al. Endocrinol (Baltimore, MD) 16, 529–540. (2004) Coexpression of Cux-1 and notch 102. Clark, A., Wilson, M., London, N. et al. signaling pathway components during kidney (1995) Identification and characterization of development. Dev Dyn 231(4), 828–838. a functional retinoic acid/thyroid hormone- 105. López-Ríos, J., Tessmar, K., Loosli F. et al. response element upstream of the human (2003) Six3 and Six6 is moduated by mem- insulin gene enhancer. Biochem J 309, bers of the groucho family. Development 130, 863–870. 185–195.